1 // SPDX-License-Identifier: GPL-2.0
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
16 #include "btree_journal_iter.h"
17 #include "btree_key_cache.h"
18 #include "btree_update_interior.h"
20 #include "btree_write_buffer.h"
21 #include "buckets_waiting_for_journal.h"
28 #include "disk_groups.h"
34 #include "fs-io-buffered.h"
35 #include "fs-io-direct.h"
41 #include "journal_reclaim.h"
42 #include "journal_seq_blacklist.h"
46 #include "nocow_locking.h"
48 #include "rebalance.h"
52 #include "sb-errors.h"
53 #include "sb-members.h"
55 #include "subvolume.h"
61 #include <linux/backing-dev.h>
62 #include <linux/blkdev.h>
63 #include <linux/debugfs.h>
64 #include <linux/device.h>
65 #include <linux/idr.h>
66 #include <linux/module.h>
67 #include <linux/percpu.h>
68 #include <linux/random.h>
69 #include <linux/sysfs.h>
70 #include <crypto/hash.h>
72 MODULE_LICENSE("GPL");
73 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
74 MODULE_DESCRIPTION("bcachefs filesystem");
77 static const struct attribute_group type ## _group = { \
78 .attrs = type ## _files \
81 static const struct attribute_group *type ## _groups[] = { \
86 static const struct kobj_type type ## _ktype = { \
87 .release = type ## _release, \
88 .sysfs_ops = &type ## _sysfs_ops, \
89 .default_groups = type ## _groups \
92 static void bch2_fs_release(struct kobject *);
93 static void bch2_dev_release(struct kobject *);
94 static void bch2_fs_counters_release(struct kobject *k)
98 static void bch2_fs_internal_release(struct kobject *k)
102 static void bch2_fs_opts_dir_release(struct kobject *k)
106 static void bch2_fs_time_stats_release(struct kobject *k)
111 KTYPE(bch2_fs_counters);
112 KTYPE(bch2_fs_internal);
113 KTYPE(bch2_fs_opts_dir);
114 KTYPE(bch2_fs_time_stats);
117 static struct kset *bcachefs_kset;
118 static LIST_HEAD(bch_fs_list);
119 static DEFINE_MUTEX(bch_fs_list_lock);
121 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
123 static void bch2_dev_free(struct bch_dev *);
124 static int bch2_dev_alloc(struct bch_fs *, unsigned);
125 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
126 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
128 struct bch_fs *bch2_dev_to_fs(dev_t dev)
134 mutex_lock(&bch_fs_list_lock);
137 list_for_each_entry(c, &bch_fs_list, list)
138 for_each_member_device_rcu(ca, c, i, NULL)
139 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
146 mutex_unlock(&bch_fs_list_lock);
151 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
155 lockdep_assert_held(&bch_fs_list_lock);
157 list_for_each_entry(c, &bch_fs_list, list)
158 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
164 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
168 mutex_lock(&bch_fs_list_lock);
169 c = __bch2_uuid_to_fs(uuid);
172 mutex_unlock(&bch_fs_list_lock);
177 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
180 unsigned i, nr = 0, u64s =
181 ((sizeof(struct jset_entry_dev_usage) +
182 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
186 for_each_member_device_rcu(ca, c, i, NULL)
190 bch2_journal_entry_res_resize(&c->journal,
191 &c->dev_usage_journal_res, u64s * nr);
194 /* Filesystem RO/RW: */
197 * For startup/shutdown of RW stuff, the dependencies are:
199 * - foreground writes depend on copygc and rebalance (to free up space)
201 * - copygc and rebalance depend on mark and sweep gc (they actually probably
202 * don't because they either reserve ahead of time or don't block if
203 * allocations fail, but allocations can require mark and sweep gc to run
204 * because of generation number wraparound)
206 * - all of the above depends on the allocator threads
208 * - allocator depends on the journal (when it rewrites prios and gens)
211 static void __bch2_fs_read_only(struct bch_fs *c)
214 unsigned i, clean_passes = 0;
218 bch2_open_buckets_stop(c, NULL, true);
219 bch2_rebalance_stop(c);
221 bch2_gc_thread_stop(c);
224 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
225 journal_cur_seq(&c->journal));
230 if (bch2_btree_interior_updates_flush(c) ||
231 bch2_journal_flush_all_pins(&c->journal) ||
232 bch2_btree_flush_all_writes(c) ||
233 seq != atomic64_read(&c->journal.seq)) {
234 seq = atomic64_read(&c->journal.seq);
237 } while (clean_passes < 2);
239 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
240 journal_cur_seq(&c->journal));
242 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
243 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
244 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
245 bch2_fs_journal_stop(&c->journal);
248 * After stopping journal:
250 for_each_member_device(ca, c, i)
251 bch2_dev_allocator_remove(c, ca);
254 #ifndef BCH_WRITE_REF_DEBUG
255 static void bch2_writes_disabled(struct percpu_ref *writes)
257 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
259 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
260 wake_up(&bch2_read_only_wait);
264 void bch2_fs_read_only(struct bch_fs *c)
266 if (!test_bit(BCH_FS_RW, &c->flags)) {
267 bch2_journal_reclaim_stop(&c->journal);
271 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
274 * Block new foreground-end write operations from starting - any new
275 * writes will return -EROFS:
277 set_bit(BCH_FS_GOING_RO, &c->flags);
278 #ifndef BCH_WRITE_REF_DEBUG
279 percpu_ref_kill(&c->writes);
281 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
282 bch2_write_ref_put(c, i);
286 * If we're not doing an emergency shutdown, we want to wait on
287 * outstanding writes to complete so they don't see spurious errors due
288 * to shutting down the allocator:
290 * If we are doing an emergency shutdown outstanding writes may
291 * hang until we shutdown the allocator so we don't want to wait
292 * on outstanding writes before shutting everything down - but
293 * we do need to wait on them before returning and signalling
294 * that going RO is complete:
296 wait_event(bch2_read_only_wait,
297 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
298 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
300 __bch2_fs_read_only(c);
302 wait_event(bch2_read_only_wait,
303 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
305 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
306 clear_bit(BCH_FS_GOING_RO, &c->flags);
308 if (!bch2_journal_error(&c->journal) &&
309 !test_bit(BCH_FS_ERROR, &c->flags) &&
310 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
311 test_bit(BCH_FS_STARTED, &c->flags) &&
312 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
313 !c->opts.norecovery) {
314 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
315 BUG_ON(atomic_read(&c->btree_cache.dirty));
316 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
317 BUG_ON(c->btree_write_buffer.inc.keys.nr);
318 BUG_ON(c->btree_write_buffer.flushing.keys.nr);
320 bch_verbose(c, "marking filesystem clean");
321 bch2_fs_mark_clean(c);
324 clear_bit(BCH_FS_RW, &c->flags);
327 static void bch2_fs_read_only_work(struct work_struct *work)
330 container_of(work, struct bch_fs, read_only_work);
332 down_write(&c->state_lock);
333 bch2_fs_read_only(c);
334 up_write(&c->state_lock);
337 static void bch2_fs_read_only_async(struct bch_fs *c)
339 queue_work(system_long_wq, &c->read_only_work);
342 bool bch2_fs_emergency_read_only(struct bch_fs *c)
344 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
346 bch2_journal_halt(&c->journal);
347 bch2_fs_read_only_async(c);
349 wake_up(&bch2_read_only_wait);
353 static int bch2_fs_read_write_late(struct bch_fs *c)
358 * Data move operations can't run until after check_snapshots has
359 * completed, and bch2_snapshot_is_ancestor() is available.
361 * Ideally we'd start copygc/rebalance earlier instead of waiting for
362 * all of recovery/fsck to complete:
364 ret = bch2_copygc_start(c);
366 bch_err(c, "error starting copygc thread");
370 ret = bch2_rebalance_start(c);
372 bch_err(c, "error starting rebalance thread");
379 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
385 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
386 bch_err(c, "cannot go rw, unfixed btree errors");
387 return -BCH_ERR_erofs_unfixed_errors;
390 if (test_bit(BCH_FS_RW, &c->flags))
393 if (c->opts.norecovery)
394 return -BCH_ERR_erofs_norecovery;
397 * nochanges is used for fsck -n mode - we have to allow going rw
398 * during recovery for that to work:
400 if (c->opts.nochanges && (!early || c->opts.read_only))
401 return -BCH_ERR_erofs_nochanges;
403 bch_info(c, "going read-write");
405 ret = bch2_sb_members_v2_init(c);
409 ret = bch2_fs_mark_dirty(c);
413 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
416 * First journal write must be a flush write: after a clean shutdown we
417 * don't read the journal, so the first journal write may end up
418 * overwriting whatever was there previously, and there must always be
419 * at least one non-flush write in the journal or recovery will fail:
421 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
423 for_each_rw_member(ca, c, i)
424 bch2_dev_allocator_add(c, ca);
425 bch2_recalc_capacity(c);
427 ret = bch2_gc_thread_start(c);
429 bch_err(c, "error starting gc thread");
433 ret = bch2_journal_reclaim_start(&c->journal);
438 ret = bch2_fs_read_write_late(c);
443 #ifndef BCH_WRITE_REF_DEBUG
444 percpu_ref_reinit(&c->writes);
446 for (i = 0; i < BCH_WRITE_REF_NR; i++) {
447 BUG_ON(atomic_long_read(&c->writes[i]));
448 atomic_long_inc(&c->writes[i]);
451 set_bit(BCH_FS_RW, &c->flags);
452 set_bit(BCH_FS_WAS_RW, &c->flags);
455 bch2_do_invalidates(c);
456 bch2_do_stripe_deletes(c);
457 bch2_do_pending_node_rewrites(c);
460 __bch2_fs_read_only(c);
464 int bch2_fs_read_write(struct bch_fs *c)
466 return __bch2_fs_read_write(c, false);
469 int bch2_fs_read_write_early(struct bch_fs *c)
471 lockdep_assert_held(&c->state_lock);
473 return __bch2_fs_read_write(c, true);
476 /* Filesystem startup/shutdown: */
478 static void __bch2_fs_free(struct bch_fs *c)
482 for (i = 0; i < BCH_TIME_STAT_NR; i++)
483 bch2_time_stats_exit(&c->times[i]);
485 bch2_free_pending_node_rewrites(c);
486 bch2_fs_sb_errors_exit(c);
487 bch2_fs_counters_exit(c);
488 bch2_fs_snapshots_exit(c);
489 bch2_fs_quota_exit(c);
490 bch2_fs_fs_io_direct_exit(c);
491 bch2_fs_fs_io_buffered_exit(c);
492 bch2_fs_fsio_exit(c);
494 bch2_fs_encryption_exit(c);
495 bch2_fs_nocow_locking_exit(c);
496 bch2_fs_io_write_exit(c);
497 bch2_fs_io_read_exit(c);
498 bch2_fs_buckets_waiting_for_journal_exit(c);
499 bch2_fs_btree_interior_update_exit(c);
500 bch2_fs_btree_iter_exit(c);
501 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
502 bch2_fs_btree_cache_exit(c);
503 bch2_fs_replicas_exit(c);
504 bch2_fs_journal_exit(&c->journal);
505 bch2_io_clock_exit(&c->io_clock[WRITE]);
506 bch2_io_clock_exit(&c->io_clock[READ]);
507 bch2_fs_compress_exit(c);
508 bch2_journal_keys_put_initial(c);
509 BUG_ON(atomic_read(&c->journal_keys.ref));
510 bch2_fs_btree_write_buffer_exit(c);
511 percpu_free_rwsem(&c->mark_lock);
512 free_percpu(c->online_reserved);
514 darray_exit(&c->btree_roots_extra);
515 free_percpu(c->pcpu);
516 mempool_exit(&c->large_bkey_pool);
517 mempool_exit(&c->btree_bounce_pool);
518 bioset_exit(&c->btree_bio);
519 mempool_exit(&c->fill_iter);
520 #ifndef BCH_WRITE_REF_DEBUG
521 percpu_ref_exit(&c->writes);
523 kfree(rcu_dereference_protected(c->disk_groups, 1));
524 kfree(c->journal_seq_blacklist_table);
525 kfree(c->unused_inode_hints);
528 destroy_workqueue(c->write_ref_wq);
529 if (c->io_complete_wq)
530 destroy_workqueue(c->io_complete_wq);
532 destroy_workqueue(c->copygc_wq);
533 if (c->btree_io_complete_wq)
534 destroy_workqueue(c->btree_io_complete_wq);
535 if (c->btree_update_wq)
536 destroy_workqueue(c->btree_update_wq);
538 bch2_free_super(&c->disk_sb);
539 kvpfree(c, sizeof(*c));
540 module_put(THIS_MODULE);
543 static void bch2_fs_release(struct kobject *kobj)
545 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
550 void __bch2_fs_stop(struct bch_fs *c)
555 bch_verbose(c, "shutting down");
557 set_bit(BCH_FS_STOPPING, &c->flags);
559 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
561 down_write(&c->state_lock);
562 bch2_fs_read_only(c);
563 up_write(&c->state_lock);
565 for_each_member_device(ca, c, i)
566 if (ca->kobj.state_in_sysfs &&
568 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
570 if (c->kobj.state_in_sysfs)
571 kobject_del(&c->kobj);
573 bch2_fs_debug_exit(c);
574 bch2_fs_chardev_exit(c);
576 kobject_put(&c->counters_kobj);
577 kobject_put(&c->time_stats);
578 kobject_put(&c->opts_dir);
579 kobject_put(&c->internal);
581 /* btree prefetch might have kicked off reads in the background: */
582 bch2_btree_flush_all_reads(c);
584 for_each_member_device(ca, c, i)
585 cancel_work_sync(&ca->io_error_work);
587 cancel_work_sync(&c->read_only_work);
590 void bch2_fs_free(struct bch_fs *c)
594 mutex_lock(&bch_fs_list_lock);
596 mutex_unlock(&bch_fs_list_lock);
598 closure_sync(&c->cl);
599 closure_debug_destroy(&c->cl);
601 for (i = 0; i < c->sb.nr_devices; i++) {
602 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
605 bch2_free_super(&ca->disk_sb);
610 bch_verbose(c, "shutdown complete");
612 kobject_put(&c->kobj);
615 void bch2_fs_stop(struct bch_fs *c)
621 static int bch2_fs_online(struct bch_fs *c)
627 lockdep_assert_held(&bch_fs_list_lock);
629 if (__bch2_uuid_to_fs(c->sb.uuid)) {
630 bch_err(c, "filesystem UUID already open");
634 ret = bch2_fs_chardev_init(c);
636 bch_err(c, "error creating character device");
640 bch2_fs_debug_init(c);
642 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
643 kobject_add(&c->internal, &c->kobj, "internal") ?:
644 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
645 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
646 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
648 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
649 bch2_opts_create_sysfs_files(&c->opts_dir);
651 bch_err(c, "error creating sysfs objects");
655 down_write(&c->state_lock);
657 for_each_member_device(ca, c, i) {
658 ret = bch2_dev_sysfs_online(c, ca);
660 bch_err(c, "error creating sysfs objects");
661 percpu_ref_put(&ca->ref);
666 BUG_ON(!list_empty(&c->list));
667 list_add(&c->list, &bch_fs_list);
669 up_write(&c->state_lock);
673 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
676 struct printbuf name = PRINTBUF;
677 unsigned i, iter_size;
680 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
682 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
686 __module_get(THIS_MODULE);
688 closure_init(&c->cl, NULL);
690 c->kobj.kset = bcachefs_kset;
691 kobject_init(&c->kobj, &bch2_fs_ktype);
692 kobject_init(&c->internal, &bch2_fs_internal_ktype);
693 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
694 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
695 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
698 c->disk_sb.fs_sb = true;
700 init_rwsem(&c->state_lock);
701 mutex_init(&c->sb_lock);
702 mutex_init(&c->replicas_gc_lock);
703 mutex_init(&c->btree_root_lock);
704 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
706 init_rwsem(&c->gc_lock);
707 mutex_init(&c->gc_gens_lock);
708 atomic_set(&c->journal_keys.ref, 1);
709 c->journal_keys.initial_ref_held = true;
711 for (i = 0; i < BCH_TIME_STAT_NR; i++)
712 bch2_time_stats_init(&c->times[i]);
714 bch2_fs_copygc_init(c);
715 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
716 bch2_fs_btree_interior_update_init_early(c);
717 bch2_fs_allocator_background_init(c);
718 bch2_fs_allocator_foreground_init(c);
719 bch2_fs_rebalance_init(c);
720 bch2_fs_quota_init(c);
721 bch2_fs_ec_init_early(c);
722 bch2_fs_move_init(c);
723 bch2_fs_sb_errors_init_early(c);
725 INIT_LIST_HEAD(&c->list);
727 mutex_init(&c->usage_scratch_lock);
729 mutex_init(&c->bio_bounce_pages_lock);
730 mutex_init(&c->snapshot_table_lock);
731 init_rwsem(&c->snapshot_create_lock);
733 spin_lock_init(&c->btree_write_error_lock);
735 INIT_WORK(&c->journal_seq_blacklist_gc_work,
736 bch2_blacklist_entries_gc);
738 INIT_LIST_HEAD(&c->journal_iters);
740 INIT_LIST_HEAD(&c->fsck_error_msgs);
741 mutex_init(&c->fsck_error_msgs_lock);
743 seqcount_init(&c->gc_pos_lock);
745 seqcount_init(&c->usage_lock);
747 sema_init(&c->io_in_flight, 128);
749 INIT_LIST_HEAD(&c->vfs_inodes_list);
750 mutex_init(&c->vfs_inodes_lock);
752 c->copy_gc_enabled = 1;
753 c->rebalance.enabled = 1;
754 c->promote_whole_extents = true;
756 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
757 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
758 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
760 bch2_fs_btree_cache_init_early(&c->btree_cache);
762 mutex_init(&c->sectors_available_lock);
764 ret = percpu_init_rwsem(&c->mark_lock);
768 mutex_lock(&c->sb_lock);
769 ret = bch2_sb_to_fs(c, sb);
770 mutex_unlock(&c->sb_lock);
775 pr_uuid(&name, c->sb.user_uuid.b);
776 strscpy(c->name, name.buf, sizeof(c->name));
777 printbuf_exit(&name);
779 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
784 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
785 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
786 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
788 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
789 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
790 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
792 c->opts = bch2_opts_default;
793 ret = bch2_opts_from_sb(&c->opts, sb);
797 bch2_opts_apply(&c->opts, opts);
799 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
800 if (c->opts.inodes_use_key_cache)
801 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
802 c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
804 c->block_bits = ilog2(block_sectors(c));
805 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
807 if (bch2_fs_init_fault("fs_alloc")) {
808 bch_err(c, "fs_alloc fault injected");
813 iter_size = sizeof(struct sort_iter) +
814 (btree_blocks(c) + 1) * 2 *
815 sizeof(struct sort_iter_set);
817 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
819 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
820 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
821 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
822 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
823 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
824 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
825 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
826 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
827 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
829 #ifndef BCH_WRITE_REF_DEBUG
830 percpu_ref_init(&c->writes, bch2_writes_disabled,
831 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
833 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
834 bioset_init(&c->btree_bio, 1,
835 max(offsetof(struct btree_read_bio, bio),
836 offsetof(struct btree_write_bio, wbio.bio)),
837 BIOSET_NEED_BVECS) ||
838 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
839 !(c->online_reserved = alloc_percpu(u64)) ||
840 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
842 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
843 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
844 sizeof(u64), GFP_KERNEL))) {
845 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
849 ret = bch2_fs_counters_init(c) ?:
850 bch2_fs_sb_errors_init(c) ?:
851 bch2_io_clock_init(&c->io_clock[READ]) ?:
852 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
853 bch2_fs_journal_init(&c->journal) ?:
854 bch2_fs_replicas_init(c) ?:
855 bch2_fs_btree_cache_init(c) ?:
856 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
857 bch2_fs_btree_iter_init(c) ?:
858 bch2_fs_btree_interior_update_init(c) ?:
859 bch2_fs_buckets_waiting_for_journal_init(c) ?:
860 bch2_fs_btree_write_buffer_init(c) ?:
861 bch2_fs_subvolumes_init(c) ?:
862 bch2_fs_io_read_init(c) ?:
863 bch2_fs_io_write_init(c) ?:
864 bch2_fs_nocow_locking_init(c) ?:
865 bch2_fs_encryption_init(c) ?:
866 bch2_fs_compress_init(c) ?:
867 bch2_fs_ec_init(c) ?:
868 bch2_fs_fsio_init(c) ?:
869 bch2_fs_fs_io_buffered_init(c) ?:
870 bch2_fs_fs_io_direct_init(c);
874 for (i = 0; i < c->sb.nr_devices; i++)
875 if (bch2_dev_exists(c->disk_sb.sb, i) &&
876 bch2_dev_alloc(c, i)) {
881 bch2_journal_entry_res_resize(&c->journal,
882 &c->btree_root_journal_res,
883 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
884 bch2_dev_usage_journal_reserve(c);
885 bch2_journal_entry_res_resize(&c->journal,
886 &c->clock_journal_res,
887 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
889 mutex_lock(&bch_fs_list_lock);
890 ret = bch2_fs_online(c);
891 mutex_unlock(&bch_fs_list_lock);
904 static void print_mount_opts(struct bch_fs *c)
907 struct printbuf p = PRINTBUF;
910 prt_str(&p, "mounting version ");
911 bch2_version_to_text(&p, c->sb.version);
913 if (c->opts.read_only) {
914 prt_str(&p, " opts=");
916 prt_printf(&p, "ro");
919 for (i = 0; i < bch2_opts_nr; i++) {
920 const struct bch_option *opt = &bch2_opt_table[i];
921 u64 v = bch2_opt_get_by_id(&c->opts, i);
923 if (!(opt->flags & OPT_MOUNT))
926 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
929 prt_str(&p, first ? " opts=" : ",");
931 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
934 bch_info(c, "%s", p.buf);
938 int bch2_fs_start(struct bch_fs *c)
941 time64_t now = ktime_get_real_seconds();
947 down_write(&c->state_lock);
949 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
951 mutex_lock(&c->sb_lock);
953 ret = bch2_sb_members_v2_init(c);
955 mutex_unlock(&c->sb_lock);
959 for_each_online_member(ca, c, i)
960 bch2_members_v2_get_mut(c->disk_sb.sb, i)->last_mount = cpu_to_le64(now);
962 mutex_unlock(&c->sb_lock);
964 for_each_rw_member(ca, c, i)
965 bch2_dev_allocator_add(c, ca);
966 bch2_recalc_capacity(c);
968 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
969 ? bch2_fs_recovery(c)
970 : bch2_fs_initialize(c);
974 ret = bch2_opts_check_may_set(c);
978 if (bch2_fs_init_fault("fs_start")) {
979 bch_err(c, "fs_start fault injected");
984 set_bit(BCH_FS_STARTED, &c->flags);
986 if (c->opts.read_only || c->opts.nochanges) {
987 bch2_fs_read_only(c);
989 ret = !test_bit(BCH_FS_RW, &c->flags)
990 ? bch2_fs_read_write(c)
991 : bch2_fs_read_write_late(c);
998 up_write(&c->state_lock);
1001 bch_err_msg(c, ret, "starting filesystem");
1005 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1007 struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1009 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1010 return -BCH_ERR_mismatched_block_size;
1012 if (le16_to_cpu(m.bucket_size) <
1013 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1014 return -BCH_ERR_bucket_size_too_small;
1019 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1021 struct bch_sb *newest =
1022 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1024 if (!uuid_equal(&fs->uuid, &sb->uuid))
1025 return -BCH_ERR_device_not_a_member_of_filesystem;
1027 if (!bch2_dev_exists(newest, sb->dev_idx))
1028 return -BCH_ERR_device_has_been_removed;
1030 if (fs->block_size != sb->block_size)
1031 return -BCH_ERR_mismatched_block_size;
1036 /* Device startup/shutdown: */
1038 static void bch2_dev_release(struct kobject *kobj)
1040 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1045 static void bch2_dev_free(struct bch_dev *ca)
1047 cancel_work_sync(&ca->io_error_work);
1049 if (ca->kobj.state_in_sysfs &&
1051 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1053 if (ca->kobj.state_in_sysfs)
1054 kobject_del(&ca->kobj);
1056 bch2_free_super(&ca->disk_sb);
1057 bch2_dev_journal_exit(ca);
1059 free_percpu(ca->io_done);
1060 bioset_exit(&ca->replica_set);
1061 bch2_dev_buckets_free(ca);
1062 free_page((unsigned long) ca->sb_read_scratch);
1064 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1065 bch2_time_stats_exit(&ca->io_latency[READ]);
1067 percpu_ref_exit(&ca->io_ref);
1068 percpu_ref_exit(&ca->ref);
1069 kobject_put(&ca->kobj);
1072 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1075 lockdep_assert_held(&c->state_lock);
1077 if (percpu_ref_is_zero(&ca->io_ref))
1080 __bch2_dev_read_only(c, ca);
1082 reinit_completion(&ca->io_ref_completion);
1083 percpu_ref_kill(&ca->io_ref);
1084 wait_for_completion(&ca->io_ref_completion);
1086 if (ca->kobj.state_in_sysfs) {
1087 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1088 sysfs_remove_link(&ca->kobj, "block");
1091 bch2_free_super(&ca->disk_sb);
1092 bch2_dev_journal_exit(ca);
1095 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1097 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1099 complete(&ca->ref_completion);
1102 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1104 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1106 complete(&ca->io_ref_completion);
1109 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1113 if (!c->kobj.state_in_sysfs)
1116 if (!ca->kobj.state_in_sysfs) {
1117 ret = kobject_add(&ca->kobj, &c->kobj,
1118 "dev-%u", ca->dev_idx);
1123 if (ca->disk_sb.bdev) {
1124 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1126 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1130 ret = sysfs_create_link(&ca->kobj, block, "block");
1138 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1139 struct bch_member *member)
1144 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1148 kobject_init(&ca->kobj, &bch2_dev_ktype);
1149 init_completion(&ca->ref_completion);
1150 init_completion(&ca->io_ref_completion);
1152 init_rwsem(&ca->bucket_lock);
1154 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1156 bch2_time_stats_init(&ca->io_latency[READ]);
1157 bch2_time_stats_init(&ca->io_latency[WRITE]);
1159 ca->mi = bch2_mi_to_cpu(member);
1161 for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1162 atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1164 ca->uuid = member->uuid;
1166 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1167 ca->mi.bucket_size / btree_sectors(c));
1169 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1171 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1172 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1173 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1174 bch2_dev_buckets_alloc(c, ca) ||
1175 bioset_init(&ca->replica_set, 4,
1176 offsetof(struct bch_write_bio, bio), 0) ||
1177 !(ca->io_done = alloc_percpu(*ca->io_done)))
1186 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1189 ca->dev_idx = dev_idx;
1190 __set_bit(ca->dev_idx, ca->self.d);
1191 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1194 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1196 if (bch2_dev_sysfs_online(c, ca))
1197 pr_warn("error creating sysfs objects");
1200 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1202 struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1203 struct bch_dev *ca = NULL;
1206 if (bch2_fs_init_fault("dev_alloc"))
1209 ca = __bch2_dev_alloc(c, &member);
1215 bch2_dev_attach(c, ca, dev_idx);
1220 return -BCH_ERR_ENOMEM_dev_alloc;
1223 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1227 if (bch2_dev_is_online(ca)) {
1228 bch_err(ca, "already have device online in slot %u",
1230 return -BCH_ERR_device_already_online;
1233 if (get_capacity(sb->bdev->bd_disk) <
1234 ca->mi.bucket_size * ca->mi.nbuckets) {
1235 bch_err(ca, "cannot online: device too small");
1236 return -BCH_ERR_device_size_too_small;
1239 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1241 ret = bch2_dev_journal_init(ca, sb->sb);
1247 memset(sb, 0, sizeof(*sb));
1249 ca->dev = ca->disk_sb.bdev->bd_dev;
1251 percpu_ref_reinit(&ca->io_ref);
1256 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1261 lockdep_assert_held(&c->state_lock);
1263 if (le64_to_cpu(sb->sb->seq) >
1264 le64_to_cpu(c->disk_sb.sb->seq))
1265 bch2_sb_to_fs(c, sb->sb);
1267 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1268 !c->devs[sb->sb->dev_idx]);
1270 ca = bch_dev_locked(c, sb->sb->dev_idx);
1272 ret = __bch2_dev_attach_bdev(ca, sb);
1276 bch2_dev_sysfs_online(c, ca);
1278 if (c->sb.nr_devices == 1)
1279 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1280 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1282 rebalance_wakeup(c);
1286 /* Device management: */
1289 * Note: this function is also used by the error paths - when a particular
1290 * device sees an error, we call it to determine whether we can just set the
1291 * device RO, or - if this function returns false - we'll set the whole
1294 * XXX: maybe we should be more explicit about whether we're changing state
1295 * because we got an error or what have you?
1297 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1298 enum bch_member_state new_state, int flags)
1300 struct bch_devs_mask new_online_devs;
1301 struct bch_dev *ca2;
1302 int i, nr_rw = 0, required;
1304 lockdep_assert_held(&c->state_lock);
1306 switch (new_state) {
1307 case BCH_MEMBER_STATE_rw:
1309 case BCH_MEMBER_STATE_ro:
1310 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1313 /* do we have enough devices to write to? */
1314 for_each_member_device(ca2, c, i)
1316 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1318 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1319 ? c->opts.metadata_replicas
1320 : c->opts.metadata_replicas_required,
1321 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1322 ? c->opts.data_replicas
1323 : c->opts.data_replicas_required);
1325 return nr_rw >= required;
1326 case BCH_MEMBER_STATE_failed:
1327 case BCH_MEMBER_STATE_spare:
1328 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1329 ca->mi.state != BCH_MEMBER_STATE_ro)
1332 /* do we have enough devices to read from? */
1333 new_online_devs = bch2_online_devs(c);
1334 __clear_bit(ca->dev_idx, new_online_devs.d);
1336 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1342 static bool bch2_fs_may_start(struct bch_fs *c)
1345 unsigned i, flags = 0;
1347 if (c->opts.very_degraded)
1348 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1350 if (c->opts.degraded)
1351 flags |= BCH_FORCE_IF_DEGRADED;
1353 if (!c->opts.degraded &&
1354 !c->opts.very_degraded) {
1355 mutex_lock(&c->sb_lock);
1357 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1358 if (!bch2_dev_exists(c->disk_sb.sb, i))
1361 ca = bch_dev_locked(c, i);
1363 if (!bch2_dev_is_online(ca) &&
1364 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1365 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1366 mutex_unlock(&c->sb_lock);
1370 mutex_unlock(&c->sb_lock);
1373 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1376 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1379 * The allocator thread itself allocates btree nodes, so stop it first:
1381 bch2_dev_allocator_remove(c, ca);
1382 bch2_dev_journal_stop(&c->journal, ca);
1385 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1387 lockdep_assert_held(&c->state_lock);
1389 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1391 bch2_dev_allocator_add(c, ca);
1392 bch2_recalc_capacity(c);
1395 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1396 enum bch_member_state new_state, int flags)
1398 struct bch_member *m;
1401 if (ca->mi.state == new_state)
1404 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1405 return -BCH_ERR_device_state_not_allowed;
1407 if (new_state != BCH_MEMBER_STATE_rw)
1408 __bch2_dev_read_only(c, ca);
1410 bch_notice(ca, "%s", bch2_member_states[new_state]);
1412 mutex_lock(&c->sb_lock);
1413 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1414 SET_BCH_MEMBER_STATE(m, new_state);
1415 bch2_write_super(c);
1416 mutex_unlock(&c->sb_lock);
1418 if (new_state == BCH_MEMBER_STATE_rw)
1419 __bch2_dev_read_write(c, ca);
1421 rebalance_wakeup(c);
1426 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1427 enum bch_member_state new_state, int flags)
1431 down_write(&c->state_lock);
1432 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1433 up_write(&c->state_lock);
1438 /* Device add/removal: */
1440 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1442 struct bpos start = POS(ca->dev_idx, 0);
1443 struct bpos end = POS(ca->dev_idx, U64_MAX);
1447 * We clear the LRU and need_discard btrees first so that we don't race
1448 * with bch2_do_invalidates() and bch2_do_discards()
1450 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1451 BTREE_TRIGGER_NORUN, NULL) ?:
1452 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1453 BTREE_TRIGGER_NORUN, NULL) ?:
1454 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1455 BTREE_TRIGGER_NORUN, NULL) ?:
1456 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1457 BTREE_TRIGGER_NORUN, NULL) ?:
1458 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1459 BTREE_TRIGGER_NORUN, NULL) ?:
1460 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1461 BTREE_TRIGGER_NORUN, NULL);
1463 bch_err_msg(c, ret, "removing dev alloc info");
1468 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1470 struct bch_member *m;
1471 unsigned dev_idx = ca->dev_idx, data;
1474 down_write(&c->state_lock);
1477 * We consume a reference to ca->ref, regardless of whether we succeed
1480 percpu_ref_put(&ca->ref);
1482 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1483 bch_err(ca, "Cannot remove without losing data");
1484 ret = -BCH_ERR_device_state_not_allowed;
1488 __bch2_dev_read_only(c, ca);
1490 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1492 bch_err_msg(ca, ret, "dropping data");
1496 ret = bch2_dev_remove_alloc(c, ca);
1498 bch_err_msg(ca, ret, "deleting alloc info");
1502 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1504 bch_err_msg(ca, ret, "flushing journal");
1508 ret = bch2_journal_flush(&c->journal);
1510 bch_err(ca, "journal error");
1514 ret = bch2_replicas_gc2(c);
1516 bch_err_msg(ca, ret, "in replicas_gc2()");
1520 data = bch2_dev_has_data(c, ca);
1522 struct printbuf data_has = PRINTBUF;
1524 prt_bitflags(&data_has, bch2_data_types, data);
1525 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1526 printbuf_exit(&data_has);
1531 __bch2_dev_offline(c, ca);
1533 mutex_lock(&c->sb_lock);
1534 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1535 mutex_unlock(&c->sb_lock);
1537 percpu_ref_kill(&ca->ref);
1538 wait_for_completion(&ca->ref_completion);
1543 * At this point the device object has been removed in-core, but the
1544 * on-disk journal might still refer to the device index via sb device
1545 * usage entries. Recovery fails if it sees usage information for an
1546 * invalid device. Flush journal pins to push the back of the journal
1547 * past now invalid device index references before we update the
1548 * superblock, but after the device object has been removed so any
1549 * further journal writes elide usage info for the device.
1551 bch2_journal_flush_all_pins(&c->journal);
1554 * Free this device's slot in the bch_member array - all pointers to
1555 * this device must be gone:
1557 mutex_lock(&c->sb_lock);
1558 m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1559 memset(&m->uuid, 0, sizeof(m->uuid));
1561 bch2_write_super(c);
1563 mutex_unlock(&c->sb_lock);
1564 up_write(&c->state_lock);
1566 bch2_dev_usage_journal_reserve(c);
1569 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1570 !percpu_ref_is_zero(&ca->io_ref))
1571 __bch2_dev_read_write(c, ca);
1572 up_write(&c->state_lock);
1576 /* Add new device to running filesystem: */
1577 int bch2_dev_add(struct bch_fs *c, const char *path)
1579 struct bch_opts opts = bch2_opts_empty();
1580 struct bch_sb_handle sb;
1581 struct bch_dev *ca = NULL;
1582 struct bch_sb_field_members_v2 *mi;
1583 struct bch_member dev_mi;
1584 unsigned dev_idx, nr_devices, u64s;
1585 struct printbuf errbuf = PRINTBUF;
1586 struct printbuf label = PRINTBUF;
1589 ret = bch2_read_super(path, &opts, &sb);
1591 bch_err_msg(c, ret, "reading super");
1595 dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1597 if (BCH_MEMBER_GROUP(&dev_mi)) {
1598 bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1599 if (label.allocation_failure) {
1605 ret = bch2_dev_may_add(sb.sb, c);
1611 ca = __bch2_dev_alloc(c, &dev_mi);
1617 bch2_dev_usage_init(ca);
1619 ret = __bch2_dev_attach_bdev(ca, &sb);
1623 ret = bch2_dev_journal_alloc(ca);
1625 bch_err_msg(c, ret, "allocating journal");
1629 down_write(&c->state_lock);
1630 mutex_lock(&c->sb_lock);
1632 ret = bch2_sb_from_fs(c, ca);
1634 bch_err_msg(c, ret, "setting up new superblock");
1638 if (dynamic_fault("bcachefs:add:no_slot"))
1641 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1642 if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1645 ret = -BCH_ERR_ENOSPC_sb_members;
1646 bch_err_msg(c, ret, "setting up new superblock");
1650 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1652 mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1653 u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1654 le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1656 mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1658 ret = -BCH_ERR_ENOSPC_sb_members;
1659 bch_err_msg(c, ret, "setting up new superblock");
1662 struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1667 m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1668 c->disk_sb.sb->nr_devices = nr_devices;
1670 ca->disk_sb.sb->dev_idx = dev_idx;
1671 bch2_dev_attach(c, ca, dev_idx);
1673 if (BCH_MEMBER_GROUP(&dev_mi)) {
1674 ret = __bch2_dev_group_set(c, ca, label.buf);
1676 bch_err_msg(c, ret, "creating new label");
1681 bch2_write_super(c);
1682 mutex_unlock(&c->sb_lock);
1684 bch2_dev_usage_journal_reserve(c);
1686 ret = bch2_trans_mark_dev_sb(c, ca);
1688 bch_err_msg(ca, ret, "marking new superblock");
1692 ret = bch2_fs_freespace_init(c);
1694 bch_err_msg(ca, ret, "initializing free space");
1698 ca->new_fs_bucket_idx = 0;
1700 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1701 __bch2_dev_read_write(c, ca);
1703 up_write(&c->state_lock);
1707 mutex_unlock(&c->sb_lock);
1708 up_write(&c->state_lock);
1712 bch2_free_super(&sb);
1713 printbuf_exit(&label);
1714 printbuf_exit(&errbuf);
1717 up_write(&c->state_lock);
1722 /* Hot add existing device to running filesystem: */
1723 int bch2_dev_online(struct bch_fs *c, const char *path)
1725 struct bch_opts opts = bch2_opts_empty();
1726 struct bch_sb_handle sb = { NULL };
1731 down_write(&c->state_lock);
1733 ret = bch2_read_super(path, &opts, &sb);
1735 up_write(&c->state_lock);
1739 dev_idx = sb.sb->dev_idx;
1741 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1743 bch_err_msg(c, ret, "bringing %s online", path);
1747 ret = bch2_dev_attach_bdev(c, &sb);
1751 ca = bch_dev_locked(c, dev_idx);
1753 ret = bch2_trans_mark_dev_sb(c, ca);
1755 bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1759 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1760 __bch2_dev_read_write(c, ca);
1762 if (!ca->mi.freespace_initialized) {
1763 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1764 bch_err_msg(ca, ret, "initializing free space");
1769 if (!ca->journal.nr) {
1770 ret = bch2_dev_journal_alloc(ca);
1771 bch_err_msg(ca, ret, "allocating journal");
1776 mutex_lock(&c->sb_lock);
1777 bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1778 cpu_to_le64(ktime_get_real_seconds());
1779 bch2_write_super(c);
1780 mutex_unlock(&c->sb_lock);
1782 up_write(&c->state_lock);
1785 up_write(&c->state_lock);
1786 bch2_free_super(&sb);
1790 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1792 down_write(&c->state_lock);
1794 if (!bch2_dev_is_online(ca)) {
1795 bch_err(ca, "Already offline");
1796 up_write(&c->state_lock);
1800 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1801 bch_err(ca, "Cannot offline required disk");
1802 up_write(&c->state_lock);
1803 return -BCH_ERR_device_state_not_allowed;
1806 __bch2_dev_offline(c, ca);
1808 up_write(&c->state_lock);
1812 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1814 struct bch_member *m;
1818 down_write(&c->state_lock);
1819 old_nbuckets = ca->mi.nbuckets;
1821 if (nbuckets < ca->mi.nbuckets) {
1822 bch_err(ca, "Cannot shrink yet");
1827 if (bch2_dev_is_online(ca) &&
1828 get_capacity(ca->disk_sb.bdev->bd_disk) <
1829 ca->mi.bucket_size * nbuckets) {
1830 bch_err(ca, "New size larger than device");
1831 ret = -BCH_ERR_device_size_too_small;
1835 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1837 bch_err_msg(ca, ret, "resizing buckets");
1841 ret = bch2_trans_mark_dev_sb(c, ca);
1845 mutex_lock(&c->sb_lock);
1846 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1847 m->nbuckets = cpu_to_le64(nbuckets);
1849 bch2_write_super(c);
1850 mutex_unlock(&c->sb_lock);
1852 if (ca->mi.freespace_initialized) {
1853 ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1858 * XXX: this is all wrong transactionally - we'll be able to do
1859 * this correctly after the disk space accounting rewrite
1861 ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1864 bch2_recalc_capacity(c);
1866 up_write(&c->state_lock);
1870 /* return with ref on ca->ref: */
1871 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1877 for_each_member_device_rcu(ca, c, i, NULL)
1878 if (!strcmp(name, ca->name))
1880 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1887 /* Filesystem open: */
1889 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1890 struct bch_opts opts)
1892 DARRAY(struct bch_sb_handle) sbs = { 0 };
1893 struct bch_fs *c = NULL;
1894 struct bch_sb_handle *sb, *best = NULL;
1895 struct printbuf errbuf = PRINTBUF;
1898 if (!try_module_get(THIS_MODULE))
1899 return ERR_PTR(-ENODEV);
1906 ret = darray_make_room(&sbs, nr_devices);
1910 for (unsigned i = 0; i < nr_devices; i++) {
1911 struct bch_sb_handle sb = { NULL };
1913 ret = bch2_read_super(devices[i], &opts, &sb);
1917 BUG_ON(darray_push(&sbs, sb));
1920 darray_for_each(sbs, sb)
1921 if (!best || le64_to_cpu(sb->sb->seq) > le64_to_cpu(best->sb->seq))
1924 darray_for_each_reverse(sbs, sb) {
1925 if (sb != best && !bch2_dev_exists(best->sb, sb->sb->dev_idx)) {
1926 pr_info("%pg has been removed, skipping", sb->bdev);
1927 bch2_free_super(sb);
1928 darray_remove_item(&sbs, sb);
1933 ret = bch2_dev_in_fs(best->sb, sb->sb);
1938 c = bch2_fs_alloc(best->sb, opts);
1939 ret = PTR_ERR_OR_ZERO(c);
1943 down_write(&c->state_lock);
1944 darray_for_each(sbs, sb) {
1945 ret = bch2_dev_attach_bdev(c, sb);
1947 up_write(&c->state_lock);
1951 up_write(&c->state_lock);
1953 if (!bch2_fs_may_start(c)) {
1954 ret = -BCH_ERR_insufficient_devices_to_start;
1958 if (!c->opts.nostart) {
1959 ret = bch2_fs_start(c);
1964 darray_for_each(sbs, sb)
1965 bch2_free_super(sb);
1967 printbuf_exit(&errbuf);
1968 module_put(THIS_MODULE);
1971 pr_err("bch_fs_open err opening %s: %s",
1972 devices[0], bch2_err_str(ret));
1974 if (!IS_ERR_OR_NULL(c))
1980 /* Global interfaces/init */
1982 static void bcachefs_exit(void)
1986 bch2_chardev_exit();
1987 bch2_btree_key_cache_exit();
1989 kset_unregister(bcachefs_kset);
1992 static int __init bcachefs_init(void)
1994 bch2_bkey_pack_test();
1996 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1997 bch2_btree_key_cache_init() ||
1998 bch2_chardev_init() ||
2009 #define BCH_DEBUG_PARAM(name, description) \
2011 module_param_named(name, bch2_##name, bool, 0644); \
2012 MODULE_PARM_DESC(name, description);
2014 #undef BCH_DEBUG_PARAM
2017 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2018 module_param_named(version, bch2_metadata_version, uint, 0400);
2020 module_exit(bcachefs_exit);
2021 module_init(bcachefs_init);